docs(cli): trim skill-shell injection comment

This commit is contained in:
Bruno Agatao
2026-08-03 13:19:03 +02:00
parent dbf50aa821
commit 87eaf2ca7d
@@ -4,15 +4,13 @@ import { Process } from "@/util/process"
import { SKILL_SHELL_DISABLED, SKILL_SHELL_UNTRUSTED } from "@/kilocode/skills/display"
import type * as Tool from "@/tool/tool"
// Shell injection for skill bodies mirrors Claude's "dynamic context injection": a
// `!`cmd`` placeholder in SKILL.md is replaced by the command's stdout before the
// A `!`cmd`` placeholder in SKILL.md is replaced by the command's stdout before the
// content reaches the model. Runs only for model-initiated skill loads (not the
// user-initiated `/skill` path), gated by: trust (only global/builtin skills, never
// project/downloaded ones), a kill-switch (KILO_DISABLE_SKILL_SHELL), and one batch
// bash permission ask naming every command up front (`skillShell` forces the ask past
// any allow/auto-approve rule; a preceding external_directory ask covers out-of-project
// paths). Substitution runs once; output is never re-scanned, so a command can't emit
// a placeholder that a later pass would execute.
// user-initiated `/skill` path), gated by trust (only global/builtin skills), a
// kill-switch (KILO_DISABLE_SKILL_SHELL), and one batch bash ask naming every command
// up front (`skillShell` forces the ask past allow/auto-approve rules; a preceding
// external_directory ask covers out-of-project paths). Substitution runs once; output
// is never re-scanned, so a command can't emit a placeholder a later pass would run.
// Execution bounds: model-initiated commands must not hang the load, blow up
// context, or overrun the batch.